Experiment-confirmed machine learning optimization of NADES extraction with ultrasound assistance and Mendelian randomization evidence on antioxidant effects of ferulic acid and ligustilide from Angelicae Sinensis Radix

This study developed an efficient and sustainable extraction process for extracting ferulic acid (FA) and ligustilide (LIG) from Angelica sinensis Radix (ASR) using green natural deep eutectic solvents (NADESs) by machine learning optimization. After screening, L-proline/methylurea was identified as the optimal NADES system, and the genetic algorithm optimized back-propagation neural network (GA-BPNN) model demonstrated superior predictive performance over traditional methods (MSE = 1.7 × 10 5 , R 2  = 0.9860). Under the optimized conditions (NADES volume fraction 67.44%, ultrasonic time 25.556 min, ultrasonic temperature 59.791 ℃, and solid–liquid ratio 7.610 mg/mL), the predicted extraction yield was 0.149, which closely matched the experimental yield of 0.151. Shapley Additive Explanations (SHAP) analysis further indicated that the NADES volume fraction was the most critical factor affecting extraction efficiency. Animals’ experiments demonstrated that both FA and LIG significantly ameliorated neurological deficits and reduced cerebral infarct volume in middle cerebral artery occlusion/reperfusion (MCAO/R) mice. To further elucidate the underlying antioxidant mechanism, Mendelian randomization (MR) analysis was employed, revealing a significant causal relationship between FA and superoxide dismutase (SOD) expression (P = 0.0265), but no significant association with heme oxygenase-1 (HO-1); LIG showed no significant association with either. These findings were corroborated in MCAO/R mice, FA administration significantly increased SOD levels in plasma. Collectively, the GA-BPNN-guided NADES combined with ultrasound assisted extraction (UAE) strategy significantly enhanced the extraction of bioactive compounds from ASR. Moreover, the integrated approach combining MR analysis with in vivo experiments provided converging evidence supporting the underlying specific antioxidant mechanism mediated by FA. Download: Download high-res image (169KB) Download: Download full-size image This study developed an efficient extraction process for ASR using NADES combined with ultrasonic assistance, optimized via machine learning models, and evaluated the antioxidant activities of FA and LIG in relation to SOD and HO-1 expression.